configuration_quasar.py

# coding=utf-8
# Copyright 2024 AstraMind and the HuggingFace Inc. team. All rights reserved.

""" Quasar model configuration"""


from transformers.configuration_utils import PretrainedConfig
from transformers.utils import logging


logger = logging.get_logger(__name__)

QUASAR_PRETRAINED_CONFIG_ARCHIVE_MAP = {
    "AstraMindAI/AstraQuasar-4B": "https://huggingface.co/AstraMindAI/AstraQuasar-4B/resolve/main/config.json",
}

#from microsoft/phi-2, Phi -> Quasar
class QuasarConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`QuasarModel`]. It is used to instantiate an Quasar
    model according to the specified arguments, defining the model architecture.

    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
    documentation from [`PretrainedConfig`] for more information.

    Args:
        vocab_size (`int`, *optional*, defaults to 51200):
            Vocabulary size of the Quasar model. Defines the number of different tokens that can be represented by the
            `inputs_ids` passed when calling [`QuasarModel`].
        hidden_size (`int`, *optional*, defaults to 2048):
            Dimension of the hidden representations.
        intermediate_size (`int`, *optional*, defaults to 8192):
            Dimension of the MLP representations.
        num_hidden_layers (`int`, *optional*, defaults to 24):
            Number of hidden layers in the Transformer decoder.
        num_attention_heads (`int`, *optional*, defaults to 32):
            Number of attention heads for each attention layer in the Transformer decoder.
        num_key_value_heads (`int`, *optional*):
            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
            `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
            by meanpooling all the original heads within that group. For more details checkout [this
            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
            `num_attention_heads`.
        resid_pdrop (`float`, *optional*, defaults to 0.0):
            Dropout probability for mlp outputs.
        embd_pdrop (`int`, *optional*, defaults to 0.0):
            The dropout ratio for the embeddings.
        attention_dropout (`float`, *optional*, defaults to 0.0):
            The dropout ratio after computing the attention scores.
        hidden_act (`str` or `function`, *optional*, defaults to `"gelu_new"`):
            The non-linear activation function (function or string) in the decoder.
        max_position_embeddings (`int`, *optional*, defaults to 2048):
            The maximum sequence length that this model might ever be used with. Quasar-1 and Quasar-1.5 supports up to 2048
            tokens.
        initializer_range (`float`, *optional*, defaults to 0.02):
            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
        layer_norm_eps (`float`, *optional*, defaults to 1e-05):
            The epsilon used by the rms normalization layers.
        use_cache (`bool`, *optional*, defaults to `True`):
            Whether or not the model should return the last key/values attentions (not used by all models). Only
            relevant if `config.is_decoder=True`. Whether to tie weight embeddings or not.
        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
            Whether to tie weight embeddings
        rope_theta (`float`, *optional*, defaults to 10000.0):
            The base period of the RoPE embeddings.
        rope_scaling (`Dict`, *optional*):
            Dictionary containing the scaling configuration for the RoPE embeddings. Currently supports two scaling
            strategies: linear and dynamic. Their scaling factor must be an float greater than 1. The expected format
            is `{"type": strategy name, "factor": scaling factor}`. When using this flag, don't update
            `max_position_embeddings` to the expected new maximum. See the following thread for more information on how
            these scaling strategies behave:
            https://www.reddit.com/r/LocalPersimmon/comments/14mrgpr/dynamically_scaled_rope_further_increases/. This
            is an experimental feature, subject to breaking API changes in future versions.
        partial_rotary_factor (`float`, *optional*, defaults to 0.5):
            Percentage of the query and keys which will have rotary embedding.
        qk_layernorm (`bool`, *optional*, defaults to `False`):
            Whether or not to normalize the Queries and Keys after projecting the hidden states.
        bos_token_id (`int`, *optional*, defaults to 1):
            Denotes beginning of sequences token id.
        eos_token_id (`int`, *optional*, defaults to 2):
            Denotes end of sequences token id.
        duplicate_trick (`bool`, *optional*, defaults to `True`):
            Whether to use the trick of self layers calling
        duplicate_grad (`bool`, *optional*, defaults to `True`):
            Whether or not to do a double grad step during training. Thi is not compatible with Gradient Checkpointing 
        remove_ff_bias (`bool`, *optional*, defaults to `True`):
            Whether or not to remove feed forward bias
        gated_activation (`bool`, *optional*, defaults to `False`):
            Whether or not to use a GeluGLU Activation
        simple_norm (`bool`, *optional*, defaults to `False`):
            Whether or not to use a simpler version of RMS Layer Norm
        sliding_window ('int', *optional* defaults to 2048):
            If specified it enables a sliding context window to extend the moel context from 2048 to 32K 
    Example:

    ```python
    >>> from transformers import AutoModel, AutoConfig
    

    >>> # Initializing a Quasar style configuration
    >>> configuration = AutoConfig.from_pretrained("AstraMindAI/AstraQuasar-4B")

    >>> # Initializing a model from the configuration
    >>> model = QuasarModel(configuration, trust_remote_code=True)

    >>> # Accessing the model configuration
    >>> configuration = model.config
    ```"""

    model_type = "quasar"
    keys_to_ignore_at_inference = ["past_key_values"]

    def __init__(
        self,
        vocab_size=51200,
        hidden_size=2560,
        intermediate_size=8192,
        num_hidden_layers=24,
        num_attention_heads=32,
        num_key_value_heads=None,
        resid_pdrop=0.0,
        embd_pdrop=0.0,
        attention_dropout=0.0,
        hidden_act="gelu_new",
        max_position_embeddings=32768,
        initializer_range=0.02,
        layer_norm_eps=1e-5,
        use_cache=True,
        tie_word_embeddings=False,
        rope_theta=10000.0,
        rope_scaling=None,
        partial_rotary_factor=0.5,
        qk_layernorm=False,
        bos_token_id=1,
        eos_token_id=2,
        sliding_window=2048,
        simple_norm=False,
        remove_ff_bias=True,
        gated_activation=False,
        duplicate_trick=True,
        duplicate_grad=True,
        layer_ranges=[[0, 16],[8, 21],[12, 25],[16, 29],[25, 32]],
        **kwargs,
    ):
        
        self.sliding_window = sliding_window
        self.simple_norm = simple_norm
        self.remove_ff_bias = remove_ff_bias
        self.gated_activation = gated_activation
        self.duplicate_trick = duplicate_trick
        self.duplicate_grad = duplicate_grad
        self.layer_ranges = layer_ranges if layer_ranges is not None else []

        self.vocab_size = vocab_size
        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads

        if num_key_value_heads is None:
            num_key_value_heads = num_attention_heads

        self.num_key_value_heads = num_key_value_heads
        self.resid_pdrop = resid_pdrop
        self.embd_pdrop = embd_pdrop
        self.attention_dropout = attention_dropout
        self.hidden_act = hidden_act
        self.max_position_embeddings = max_position_embeddings
        self.initializer_range = initializer_range
        self.layer_norm_eps = layer_norm_eps
        self.use_cache = use_cache
        self.rope_theta = rope_theta
        self.rope_scaling = rope_scaling
        self.partial_rotary_factor = partial_rotary_factor
        self.qk_layernorm = qk_layernorm
        self._rope_scaling_validation()

        super().__init__(
            bos_token_id=bos_token_id,
            eos_token_id=eos_token_id,
            tie_word_embeddings=tie_word_embeddings,
            **kwargs,
        )

    # Copied from transformers.models.llama.configuration_llama.LlamaConfig._rope_scaling_validation
    def _rope_scaling_validation(self):
        """
        Validate the `rope_scaling` configuration.
        """
        if self.rope_scaling is None:
            return

        if not isinstance(self.rope_scaling, dict) or len(self.rope_scaling) != 2:
            raise ValueError(
                "`rope_scaling` must be a dictionary with with two fields, `type` and `factor`, "
                f"got {self.rope_scaling}"
            )
        rope_scaling_type = self.rope_scaling.get("type", None)
        rope_scaling_factor = self.rope_scaling.get("factor", None)
        if rope_scaling_type is None or rope_scaling_type not in ["linear", "dynamic"]:
            raise ValueError(
                f"`rope_scaling`'s type field must be one of ['linear', 'dynamic'], got {rope_scaling_type}"
            )
        if rope_scaling_factor is None or not isinstance(rope_scaling_factor, float) or rope_scaling_factor <= 1.0:
            raise ValueError(f"`rope_scaling`'s factor field must be a float > 1, got {rope_scaling_factor}")